EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Prediction of child mortality in Tanzania: Machine learning model approach

Julius Moinget Loibor, Manash Pratim Barman, Baby Das

Clinical Epidemiology and Global Health · 2026-09 · 41 · 102442

DOI: 10.1016/j.cegh.2026.102442 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Clinical Epidemiology and Global Health, 2026-09. Verfasst von Julius Moinget Loibor und 2 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Problem considered: Predicting child mortality in Tanzania is challenged by the class imbalance in the Demographic and Health Survey (DHS) dataset, which compromises the predictive performance and interpretability of machine learning models intended to support policy decision-making. In this study, the imbalanced DHS dataset was balanced using an appropriate machine learning-based resampling technique to improve child mortality prediction.

Methods

Seven machine learning models, such as Random Forest, Support Vector Machine, Decision Tree, Linear Discriminant Analysis, Logistic Regression, LightGBM, and CatBoost, were evaluated for predictive performance before and after applying SMOTE for data balancing.

Results

After applying SMOTE, the Random Forest model outperformed other machine learning models for predicting child mortality in Tanzania, achieving an area under the curve (AUC) of 94%, an accuracy of 90%, a recall of 0.86 and an F1-Score of 0.86. Model interpretability was achieved by using the SHAP technique, which confirmed that the top predictors included total children ever born, mother's age group, Zone, wealth index, current marital status, and mother's level of education. The SHAP analysis shows that the most influential feature among the identified predictors was total children ever born. This variable should be used as a meaningful predictor in the machine learning framework, but not as a cause of child mortality, due to its potential endogeneity.

Conclusion

The SMOTE-balanced Random Forest model with SHAP provides a superior explainable framework for child mortality prediction, offering policymakers and stakeholders in Tanzania evidence-based insights to target key important predictors and strengthen healthcare interventions.

Zitieren

APA

Loibor, J. M., Barman, M. P., & Das, B. (2026). Prediction of child mortality in Tanzania: Machine learning model approach. Clinical Epidemiology and Global Health, 41, 102442. https://doi.org/10.1016/j.cegh.2026.102442

Vancouver

Loibor JM, Barman MP, Das B. Prediction of child mortality in Tanzania: Machine learning model approach. Clinical Epidemiology and Global Health. 2026;41:102442. doi:10.1016/j.cegh.2026.102442

BibTeX
@article{Loibor2026,
  author = {Loibor, Julius Moinget and Barman, Manash Pratim and Das, Baby},
  title = {Prediction of child mortality in Tanzania: Machine learning model approach},
  journal = {Clinical Epidemiology and Global Health},
  year = {2026},
  volume = {41},
  pages = {102442},
  doi = {10.1016/j.cegh.2026.102442}
}
Kostenlos starten →